Compressive Sensing Depth Video Coding via Gaussian Mixture Models and Object Edges

In this paper, we propose a novel compressive sensing depth video (CSDV) coding scheme based on Gaussian mixture models (GMM) and object edges. We first compress several depth videos to get CSDV frames in the temporal direction. A whole CSDV frame is divided into a set of non-overlap patches in which object edges is detected by Canny operator to reduce the computational complexity of quantization. Then, we allocate variable bits for different patches based on the percentages of non-zero pixels in every patch. The GMM is used to model the CSDV frame patches and design product vector quantizers to quantize CSDV frames. The experimental results show that our compression scheme achieves a significant Bjontegaard Delta (BD)-PSNR improvement about 2–10 dB when compared to the standard video coding schemes, e.g. Uniform Scalar Quantization-Differential Pulse Code Modulation (USQ-DPCM) and H.265/HEVC.